Install error with 0.3.2 about dscribe HOT 10 CLOSED

Hi @sirmarcel,

I cannot reproduce this issue on a clean python3.7 install using pip globally or within a virtualenv.

Both in this issue and #43 a conda environment is used. Maybe this has something to do with it? I will take a look into this. Could you give a try by using virtualenv instead of a conda environment?

from dscribe.

Hi -- okay, I will look into this. It might take a little way, as I'm knee-deep in a number of projects at the moment!

from dscribe.

I can reproduce the issue when installing dscribe with conda from conda-forge (i.e. conda install dscribe). The issue is however limited to the conda package manager, and is not present when installing with pip. The easiest work-around for now is to install dscribe with pip instead. You can do this even inside a conda environment with:

conda install pip
pip install pybind11
pip install dscribe

I would suggest using this for now until I can figure out the issue with the conda distribution.

from dscribe.

The issue is now also fixed in the conda package (there was actually an issue with the conda build system itself, not with dscribe). Just update to the latest version from conda-forge.

from dscribe.

Hi Lauri, that's great! In my case I was installing it via poetry though (which should just be pip under the hood), but inside a conda environment.

from dscribe.

Oh, my bad. I just saw conda in your filepath and leaped to an assumption 😛

From the error, I would guess that this might an issue with your compiler. Which version of GCC are you using? Maybe simply removing the package completely (make sure that that build/temp.linux-x86_64-3.7/dscribe/libacsf/acsfwrapper.o is removed) and switching to another compiler version could work?

from dscribe.

Hi! No worries. :D

I tried it with gcc6 and gcc9, the original report was with gcc8. It always looks like the same error. I'm not 100% sure I'm cleaning up after pip correctly, but I deleted ~/.caches/pip after each attempt, and as far as I know, pip automatically cleans up if the build fails...

from dscribe.

I'm quite sure that this has something to do with the platform or libraries in your HPC environment. What linux distribution is your HPC running? Can you attach the output of pip list within your conda env? If you are using slurm, can you attach the output of module list?

from dscribe.

Closing this for now as inactive. Reopen as needed. Also, there is a new version (0.3.5) that you may try out (the wrapper code has been updated so maybe that helps?)

from dscribe.

Hello, just a quick note: This error still occurred in my original setup even with 0.3.5, but did not persist when I did the following:

• Clean conda environment with python 3.8, numpy, scipy
• pip install dscribe

I think the original problem might have been due to the awkward interaction of having a virtualenv inside a conda environment, or something like that. Still a bit mysterious, but pleased that it seems to work.

EDIT: Same thing applies to python 3.7, still works.

from dscribe.